actix-web
actix-web copied to clipboard
Overlapping scopes work differently than flat overlapping routes
Expected Behavior
The docs seem to suggest that the web::scope(...) is a helper that groups different routes under the same prefix, but that otherwise the behaviour would be the same as with flattened routes. At least, I didn't see any mention of differences.
So, when I refactored my code from flat routes like in this minimal example:
use actix_web::{get, put, web, App, HttpServer, Responder};
#[get("/camera/dimensions")]
async fn camera_dimensions() -> impl Responder {
"800x600"
}
#[get("/camera/owner")]
async fn camera_owner() -> impl Responder {
"John Doe"
}
#[get("/focuser/distance")]
async fn focuser_focus() -> impl Responder {
"42"
}
#[put("/focuser/distance")]
async fn focuser_focus_set(value: web::Path<f64>) -> impl Responder {
format!("Set focus to {}", value)
}
#[get("/{device_type}/connected")]
async fn device_connected(device_type: web::Path<String>) -> impl Responder {
format!("{} is connected", device_type)
}
#[put("/{device_type}/connect")]
async fn device_connect(device_type: web::Path<String>) -> impl Responder {
format!("Connecting {}", device_type)
}
#[actix_web::main]
async fn main() -> std::io::Result<()> {
HttpServer::new(|| {
App::new()
.service(camera_dimensions)
.service(camera_owner)
.service(focuser_focus)
.service(focuser_focus_set)
.service(device_connected)
.service(device_connect)
})
.bind(("127.0.0.1", 8080))?
.run()
.await
}
to scoped groups like in this one:
use actix_web::{get, put, web, App, HttpServer, Responder};
#[get("/dimensions")]
async fn camera_dimensions() -> impl Responder {
"800x600"
}
#[get("/owner")]
async fn camera_owner() -> impl Responder {
"John Doe"
}
#[get("/distance")]
async fn focuser_focus() -> impl Responder {
"42"
}
#[put("/distance")]
async fn focuser_focus_set(value: web::Path<f64>) -> impl Responder {
format!("Set focus to {}", value)
}
#[get("/connected")]
async fn device_connected(device_type: web::Path<String>) -> impl Responder {
format!("{} is connected", device_type)
}
#[put("/connect")]
async fn device_connect(device_type: web::Path<String>) -> impl Responder {
format!("Connecting {}", device_type)
}
#[actix_web::main]
async fn main() -> std::io::Result<()> {
HttpServer::new(|| {
App::new()
.service(
web::scope("/camera")
.service(camera_dimensions)
.service(camera_owner),
)
.service(
web::scope("/focuser")
.service(focuser_focus)
.service(focuser_focus_set),
)
.service(
web::scope("/{device_type}")
.service(device_connected)
.service(device_connect),
)
})
.bind(("127.0.0.1", 8080))?
.run()
.await
}
I expected all routes to continue working like before.
Current Behavior
Unfortunately, it seems that scopes are only matched against each other, and, once a scope is matched, no other scopes are tried even if it doesn't have the required method.
So, routes like /camera/connected and /focuser/connected worked before, but stop working after refactoring.
I thought reordering might help, putting the generic /{device_type} scope first, but that only seems to change which scope catches everything - now, routes like /*/connected work but specific device routes like /camera/dimensions don't.
Finally, I thought maybe I should use /{device_type} as app.default_service(), but that doesn't work either as default_service() seems to only accept a single handler and not a scope as its argument.
Possible Solution
Flatten all routes at the app level when scopes are registered, so that there is no observable change in behaviour after grouping.
Steps to Reproduce (for bugs)
Run the minimal examples provided above with suggested requests like http://localhost:8080/camera/connected and observe the working response "camera is connected" in the first implementation and 404 error in the 2nd one.
Context
I'm trying to refactor my complex list of routes into grouped scopes in separate functions to make code a bit more maintainable.
Your Environment
- Rust Version (I.e, output of
rustc -V): rustc 1.63.0 (4b91a6ea7 2022-08-08) - Actix Web Version: 4.2.1
As you've noticed, Scopes match paths in a fundamentally different way to a flat route structure. In general, once a service has been matched the router will never consider services registered after it. In your case the router would never match /camera/connected to the first two routes unless using Scope("/camera") because Scope captures everything with that prefix.
once a service has been matched the router will never consider services registered after it
I do believe this is worth spelling out in the documentation.
Here's a reasonably clean solution that allows you to keep the scopes and also re-use the generic device routes: https://www.rustexplorer.com/b/6kr5zk
Using Scope::configure can help segment code in to modules the first, flat case, too, while keeping its behavior.
// camera.rs
#[get("/camera/dimensions")]
async fn camera_dimensions() -> impl Responder {
"800x600"
}
#[get("/camera/owner")]
async fn camera_owner() -> impl Responder {
"John Doe"
}
fn config_camera_routes(cfg: &mut web::ServiceConfig) -> {
cfg
.service(camera_dimensions)
.service(camera_owner)
}
// focuser.rs
fn config_focuser_routes(cfg: &mut web::ServiceConfig) -> { ... }
// generic.rs
fn config_generic_routes(cfg: &mut web::ServiceConfig) -> { ... }
// app.rs
App::new()
.configure(config_camera_routes)
.configure(config_focuser_routes)
.configure(config_generic_routes)
can help segment code in to modules the first
Yeah but doesn't help eliminate the prefix unfortunately (in my case there's dozens of routes under each group).
But yeah, at least clearly documenting this, or, better yet, detecting unreachable routes at init time, like some routers do, would be already good to save time for others who run into this.
I'll consider this particular issue closed when docs have satisfactory detail added though you may with to follow #414 and #2264 too.
A better way is to introduce ControlFlow type to router and every thing return from service is either Continue<Request> or Break<Result<Response, Error>>. With it user can choose to override default behaviour and fall through to routes under.
rocket use a similar type to give better routing control.
I have a similar issue I want to add to ADMIN routes jwt auth middleware, but there is web::scope conflict
HttpServer::new(move || {
let cors = Cors::permissive();
App::new().wrap(cors).service(
web::scope("api/v1")
//ADMIN
.service(
web::scope("/excursions")
.service(add_excursion)
.service(delete_excursion_by_id)
.service(update_excursion_by_id)
.service(
web::scope("/costs")
.service(add_customer_cost)
.service(delete_customer_cost_by_id)
.service(update_customer_cost_by_id)
.service(
web::scope("/types")
.service(add_customer_type)
.service(delete_customer_type_by_id)
.service(update_customer_type_by_id),
),
),
)
//CLIENT
.service(
web::scope("/excursions")
.service(get_all_excursions)
.service(get_excursion_by_id)
.service(
web::scope("/costs")
.service(get_customer_cost_by_excursion_id)
.service(
web::scope("/types")
.service(get_all_customer_type)
.service(get_customer_type_by_id),
),
),
),
)
})
.bind(("0.0.0.0", 8090))?
.run()
.await
all Client routes are not present (404)